To appear in Klaus Kaasgaard: Genres of Usability, forthcoming

Chapter 4
Beyond Formalisms:
The Art and Science of Designing Pliant Systems
A Talk with Austin Henderson and Jed Harris

ustin Henderson has a Ph.D. in Computer Science from MIT and has been in the
field of Human-Computer Interaction for more than twenty-five years. He has
built applications in several areas including manufacturing, air traffic control,
electronic mail, user interface design tools and workspace management. He has done
research and user interface architecture for Xerox and for Apple Computer. In addition
to being a principal in Rivendel Consulting, Austin is a co-founder of Pliant Research, a
research consortium exploring the theory and practice of computing systems that move
beyond the formal. Austin has been active in ACM/SIGCHI since 1983, including as
conference chair and organization chair.

ed Harris studied computational linguistics, computer science and the philosophy of

science at MIT and Stanford University. He has worked in artificial intelligence and
computer systems research for over twenty-five years, at Data General, Intel, and
Apple Computer. He was a founder of the OOPSLA conference and a co-architect of
the OpenDoc component software standard. For the last three years has been a member
of a venture fund in Silicon Valley. He is currently exploring the feasibility of new
technologies for implementing active patterns. He co-founded Pliant research with
Austin Henderson in 1997.

To appear in Klaus Kaasgaard: Genres of Usability, forthcoming

KLAUS: Id like to discuss the concepts and the framework that you present through
Pliant Research, for instance at conferences such as CHI in Pittsburgh in 99. So lets
start at the beginning. What is the problem with computing as you see it at Pliant
Research?
JED: Well, one of the problems that we have with the whole Pliant story is that there
are so many different ways to begin. It depends on the specific audience and it depends
on the kind of interests involved. Maybe for this context we see the problem as rigidity
in the existing computer infrastructure. That results in increasing tension over time
between the social reality and the computer infrastructure because social relationships
naturally evolve over time, meanings drift or are negotiated, whereas the computer
infrastructure just stays static until there is a big investment to make a change.
AUSTIN: And a related problem is that the ideology of building computer systems is
that you may talk to lots of users and if you really do your homework well you may get
all the viewpoints. But then you settle on one way of having the things in the machine.
They may have different meanings to different people, but you are still stuck with
having only a single thing in there. Its not going to move and it is also singular. So the
idea that the machine could be part of the process of dealing with the different
perspectives bashing against each other and thereby discovering meaning is not
thinkable within our current computing ideology, and as a result, not possible with our
existing technology because the machine has only one view point and doesnt support
multiple, local viewpoints.
KLAUS: At the CHI 99 presentation you argued that computing is very much tied to a
certain perspective on organizations or, as I would call it, a certain bureaucratic
discourse. Could you expand on that?
JED: Well, I think that its an example of how the audience shapes the way we tell the
story. That certainly is one particular way of slicing it. Another way of slicing
computing is that it is very tied to a certain view on meaning. You might say that it is
tied to a truth-value kind of meaning or a predicate calculus style of meaning. There is
another way at looking at it, as Austin says, a certain epistemology, that there is one
world and one ultimate perspective on that world that makes sense and everything else
is just a special case of that. But certainly we see those different ways of looking at it as
interrelated in a social sense. They arent just separate frameworks. And as you say, the
bureaucratic discourse is probably one of the things that we regard as the source of that.
In building organizations out of human beings, there has been an attempt to constrain
people, to limit them and to keep them operating in a consistent way in order to prevent
the organization from dissolving into chaos. That has tended to generate, in one way or
another, all these different, more rigid perspectives.
AUSTIN: You can go back and look at it historically. Im not an historian, but if you
take a look at the bureaucracies prior to 1850, they were relatively small organizations.
There might have been a reach which was larger, but still most of the management and
the organizational components of the companies all happened in one room or in one
building. Consequently management was something that could be done by a few people
bouncing off each other and a lot of it not having to be very explicit. With the arrival of

To appear in Klaus Kaasgaard: Genres of Usability, forthcoming

the telegraph and the railroad - and the reach of big companies like Joanne Yates
describes in her booki - you suddenly had a distributed organizational problem, which
can be divided into three interrelated components: size, coherence and local realities. In
certain social organizations, including large corporations, but I think much more
broadly than that, you have these three components: You have the need to respond to
local circumstances, to fit in and to couple with what is going on wherever the local
presence of the company is; you have the need to keep those separate vectors coherent;
and you have to do this in the face of getting big.
The way that organizations tended to deal with this prior to computers was by the
invention of scientific management and the idea of having rigid, well-defined file
systems and things like memos, which have particular forms and a bunch of other things
which took a lot of inventing. In that sense modern organizations were invented
between 1850 and 1920, as Yates describes. So what they had done was to get this
single file system, these single answers lined up, all marching in exactly the same
direction. They built this structure out of people. And suddenly with the arrival of the
computer in the late fifties and early sixties there was a realization that here was just the
tool they needed for keeping everything exactly aligned. That was the inevitable
response to trying to deal with this.
Now, maybe you had to have only a single way of doing things when this management
structure was being constructed entirely out of people, but the interesting thing to us is
the thought that the computer could be used in a completely different way. The
computer can be used to enable more than one view, to handle the local circumstances
and provide some of the infrastructure that will allow you to produce coherence. Rather
than enforcing a single view, you enable having more than one. That, to us, is a really
interesting thought.
JED: You also enable very dynamic negotiation. Austin has given a good overview of
what I would say is a relatively recent evolution of the bureaucratic mindset. But I really
think it goes back to the very early phases of people living in large groups, for example
Babylonian city states. And I think we can see the same kind of pattern in medieval
theology. For example, as Austin says, there is a problem of coordinating and in
maintaining some level of coherence in a very large and distributed organization and in
medieval Europe it was the church. It certainly also goes back to the renaissance
invention of bureaucracy and straight on through in military organizations. It is also no
coincidence, for example, that the ideas of mass production and interchangeable parts
were invented by the military in the production of weapons. I think it is also no
coincidence that a lot of the computing technology was invented by wartime efforts
under the direction of the military. I think it might possibly be a coincidence, but if so it
is a startling one that our ideas of mathematics came out of large civil engineering
projects in the fertile crescent and Egypt and got consolidated into an axiomatic format.
And I think to a large extent our idea of meaning and truth comes out of theological
arguments in the middle ages. So I really do believe that this is a set of issues that
permeates our culture to a degree that we are completely unaware of. We cant even see
it. It is the fabric of our enculturated reality to such an extent that it is almost impossible
to imagine a world that is any different. And I think that Austin made a key point, which
is that we dont believe that it is possible to have a world that is any different as long as

To appear in Klaus Kaasgaard: Genres of Usability, forthcoming

organizations are fundamentally only possible through enculturating people to a certain

kind of ritualized rule-following process. But when the technology shifts sufficiently, it
may be possible to get out of that.
KLAUS: Is technology one of the most important things that keep us within this worldview, so to speak?
JED: In a sense I would say the limitations of the technology.
AUSTIN: But I think that the technology is responding to something that is fairly
deeply built into our organizational structure. I was thinking the other day about the
incredibly hard work that somebody went through to create the idea of monotheism out
of the collection of several different gods who do different things. The idea that you
would create One and that all the others would be reflections of that One is the same
notion in many ways. They could only be reasonable if there were One.
JED: And you can trace ideas like the universal laws of physics straight back to
monotheism. There is no separation at all.
KLAUS: So you are saying that the kind of computing that we have today is an
ideology rather than something intrinsic to the technology, however firmly embedded it
is today. But a critique of this position might go something like this: The force of
computers is primarily to crunch numbers and therefore also to perform and support
activities that are precise and can be easily formalized, and thus support coherence for
instance. Its not a tool that is particularly well suited for supporting some of the tasks
that humans do well pattern recognition, interpretation, empathy, situated actions etc.
So wouldnt it be better to accept that the computer is good at some things but should be
kept entirely out of matters related to things that humans do better. Do you not risk, in
spite of good intentions, to support a colonization of the life-world, as Habermas would
say, rather than creating pliant systems supporting local activities?
JED: I think that is an excellent question. First of all, as I am sure you have seen here in
Silicon Valley and as people all over the world are seeing now, Habermas project of
separating the life-world from the technical world is not feasible anymore. People are
buying books and groceries over the Internet and everybody has to interact with
computers many times during the day and its just going to keep getting more so. And
that is not driven by some conspiracy. Its driven by economic efficiency and
convenience for individuals and values like that, which people in a very diverse way are
interested in.
I think that our concern is that in some sense this attempt to separate the two and to
somehow contain technology within its own little sandbox is a counsel of despair. It is
saying: The character of technology is given, its somehow autonomous, its not
socially determined, its not socially modifiable so we just have to accept it. And we
think someone who says that owes us an argument for why that's so. That is certainly
our experience of technology today, but I think we should pry the cover up and ask why
that is our experience of technology. To take an example, the experience of
manufacturing, the experience of producing physical goods, was not an alienated

To appear in Klaus Kaasgaard: Genres of Usability, forthcoming

activity until someone came along and automated it with procedures that were borrowed
from military models. It wasnt autonomously alienating or autonomously rigid. It was
made that way by imposing a certain kind of social structure on it.
AUSTIN: I also think that in your question there was the suggestion that computers
wouldnt be good at things that people are clearly good at. I am not sure how far
computing is going to go if we change the nature of it. But I dont think that my
aspirations for a new kind of computing are all in the direction of getting the computer
to do everything that humans do. I think that computers are not in the world the same
way that people are in the world and as such they will have different resonances. But
that is not to say that the computer couldnt provide some mechanical help with aspects
of things which go on as part of the beating together of ideas. I tend to want to not go
the route of Artificial Intelligence, which says that computers have to do it all alone. My
hope is to get a prosthesis for some of these things that the computer doesnt do at all
now namely the interplay of ideas and the working out of what meanings might be.
Not that the computer would do it for you, but that the computer would do it with you in
what we call a co-productive way. This would leave to humans, if you like, those
things that are essentially better done by humans. But I think the line that we have
drawn now is far short of what the computers could do to help us deal with the richness
of ideas and meanings.
JED: I endorse what Austin says. I just want to push it even a little bit further in a
certain direction. Even if we cant change the characteristics of the computer at all in
terms of its basic rigidity, we can do a lot, with no special technical changes, to embed
the computing process into the social process in a way that gives the social activities
much more control over the local circumstances of computing. So even if you totally
accept that people have to do all the empathizing and all the creativeness and all the
pattern recognition and so on, and I think that is a very good point, we can take existing
information systems and kind of add joints to them such that people can exert much
more control over them and much more control over their work circumstances.
AUSTIN: I agree completely. If you buy the problem, then the agenda we see is twofold. The first is to use the current, rigid technology in a pliant way. That would require
a change in perspective on the part of all of those of us in the KMDs of the world to
have the computer play a different role, even if it is the same technology that we know
today. The second agenda is a much broader one: to get machines that can begin to push
that mechanical edge further forward.
KLAUS: I would like to go more into your two agendas. So lets start with the first one.
Im interested in discussing some of the roads that we can all take in order to actually
use computer systems as they exist today in a more pliant manner. Ive heard you
mention that developers can squeeze it a little here and fix it a little there and it actually
would be a lot easier to use and it would support social practices a lot better. So could
you go into some of your solutions?
AUSTIN: Well, we have different kinds of things. One of them is the observation that
one of the brilliant inventions of the paper bureaucracy was the idea of the margin. The
margin is a place on a paper form, which is designed for writing things down that are

To appear in Klaus Kaasgaard: Genres of Usability, forthcoming

outside, both physically and conceptually, the form that the system expects. The thing
about the margin is that it is connected to the form in such a way that the form carries
the stuff that goes beyond the form along with the form.
JED: So it is an unformed part of the form.
AUSTIN: Exactly so. But what was good about margins in the old paper bureaucracies
was that they were uniformly there and that there were practices that meant that you
knew where to look. You looked in the margin. Of course we have similar things now:
You look at these little yellow stickies. Thats something which has become an
institutionalized way of getting beyond the form while still being solidly within the
form. Unfortunately we havent done the same thing in our computers. Not that we
couldnt; we could easily do it. Put a field called margin, or add to every field the ability
to tag it, the ability to say: Here is some more stuff. The computer of course cant do
anything more than carry it to another human being and present that its there and that
somebody had better look at it. Its part of the form, but outside the computational
capability of the form, which addresses only what is expected in the form.
So the shift that could happen is that we just have to put those margin-like things in the
form and change all the practices where computation deals with fields that have
information associated with them such that it gets back and gets processed by a human
to to see if there is something that needs to be done. So the system asks the user whether
this marginal stuff, which is now carried electronically, makes a difference. So you
genuinely put people back into the process in a way which is scary if you were hoping
to automate it. This is a step away, philosophically, from the business of figuring it out
once and for all, and then letting the machines do it. Instead, we say: The designers of
the machines will figure out something, people using them will figure out more, and
then the users and the machines together will actually do it. This shift in the social
relationship between the computation and the world it is serving is a big change and that
will take a lot of shifting. So that is one potential solution to the problem of creating a
more pliant use of current rigid computing: Margins.
JED: All the things that we have are examples. We think they are valid examples, but
we dont have a taxonomy or some kind of architectural proposal that is somehow allencompassing. I think this first one is an especially good example, because it is so
obviously implementable. You can implement it with todays databases and in Cobol. It
is something that could have been implemented from the very beginning of information
technology. There is nothing at all demanding about it. And the fact that it wasnt done
this way even though in some sense it was prefigured by the actual practices of
bureaucrats is an indication to us that this was an ideologically dictated style of
design, not a technically dictated style.
There is a whole other category of examples, which is derived from ecology. Instead of
taking software as written and provided to the users, you give the users fragments of
software that can be copied and co-aligned into different forms and you let the users
essentially construct their own local software to do what they want. Then they pass
pieces of software around and over time this population of fragments of software acts
like an evolving ecology and adapts to demands of users. I think that maybe it is worth

To appear in Klaus Kaasgaard: Genres of Usability, forthcoming

referring back to your earlier question about human skills versus computing. Notice that
in the first case the example of margins on forms the computer is doing absolutely
no interpretation, no pattern matching (maybe some string matching but thats not even
essential), no creativity. We are not trying to make the computer any more intelligent
than it is in the most boring information processing environment. In this ecological
example that is also true. The computer is simply keeping and copying and executing
little pieces of code. There is no intelligence. We are not making it any smarter than it is
in a normal course of events.
What is maybe a little scary about the ecological example is that in order to understand
the overall development of the system, you have to do a certain kind of reasoning. This
certain kind of reasoning is not so common for designers because it is no longer
possible to predict the overall function of the system any more from the design. The
design is a substrate that new things can grow on. We have three existing examples of
that. One is the buttons that were implemented at EuroPARCii by Austin and others.
Another one is the HyperCard product that Apple provided, which established a very
flourishing ecology of little pieces of code and buttons and forms that people passed
around, copied and made new applications out of. A third one thats familiar to just
about everybody is the World Wide Web, where you can always go look at the source
of a Web page and in fact innovations propagate through the Web very quickly. There is
this whole ecology of features and new mutations that sometimes propagate with
blinding speed to many Web pages. So I think this a very real example and it could be
driven deeper into the infrastructure. We could come up with much richer ecological
software processes.
KLAUS: I would like to discuss these solutions a bit further. And I would like to do
that by giving you a concrete example that I just now started to think of in terms of your
concept of pliant systems - or pliant use of rigid systems might be more correct.
AUSTIN: Or rigid technology, the system being the larger context of both humans
and machines.
KLAUS: Yes. I did a usability study of an electronic patient record for midwives that
we developed at KMD. Often, what I find in such studies is an incongruence between
different perspectives, e.g. between health authorities, which is one perspective, medical
research, which is another, and there are different practitioners each with their own
perspective. So we have a number of different perspectives, which sometimes are
incongruent. But this example surprised me a bit, because even amongst midwives I
found different perspectives and opinions. So, for example, the midwives would
disagree about the rigidity versus the flexibility of the system. Some of the midwives
were very unhappy with the fact that the system was constructed in such a way that, for
instance, you could not go from one page to another without having filled out certain
fields, which had to be filled out with the correct codes, not just free text for instance.
They found that it obstructed their daily work, their daily routines. They said: In my
daily work I dont have time to remember these codes. I need to get back to the woman
in labor instead of standing here figuring out the code. Other midwives were really
happy about this, because it reminded them of what they needed to do and helped
prevent them making errors. Some of them also mentioned that they were happy about

To appear in Klaus Kaasgaard: Genres of Usability, forthcoming

the fact that the computer functioned as a controller in this sense because it documented
that things had been carried out in the right order in case of patients filing lawsuits for
negligence or mistreatment. So they had different perspectives on this, which surprised
me a bit. Of course there are different reasons for the different perspectives. The
midwives that supported the rigidity were the younger midwives while the ones that
hated it were typically the older midwives. So there is something about the different
kinds of knowledge and experience that they have and there is something about their
career trajectories probably, but this is just an example of different perspectives within
the group of midwives and a lot of questions come along with this example. We can
take them one at a time. One is that you seem to assume that people want flexibility,
that people want pliant systems. How can you be sure?
AUSTIN: We assume that the world is a sufficiently rich place with enough pressures
on different people that they want to be able both to respond in the way which the world
is pushing them locally and yet at the same time to be able to be coherent with other
people who are feeling different pressures. How do we get that to happen? At the
moment, as weve said, we try to think it all out in advance and put a plan in place and
then everybody marches to that drum. So people will say: Well all agree to fill out
certain fields before we change pages, or theyll say: Well all agree that you can ask
later or maybe theyll consider a way to get some alerts. These are all design
considerations that are thought out in advance. You might even have a switch such that
you can individually customize it, but you are still thinking them all out in advance.
The thing that we are assuming that people want is not flexibility for its own sake. But
they do want to be able to do their own thing. As the world changes and they suddenly
see things differently, they want to be able to act accordingly. For example, a midwife
will say: Oh these ones I know about, but this field is particularly critical, I want an
alert for me with respect to that. If you thought it all out in advance either it flips the
page or it doesnt. Then you dont have that thing that maybe the logic in the workflow
is differential, maybe its more situated in the sense that its based on which question is
asked, when, by whom and how. In fact all those possibilities just enumerate forever.
There are a million different circumstances so we cant plan the questions and answers
in advance. The richness of life is such that, in the particular circumstance, I want to be
able to say: Well, the pattern for me is this and then have the system be able to help at
some level. So I am saying: Do they want that flexibility for its own sake? No! They
want it in response to their need and what they notice about themselves and the way
they work.
KLAUS: But how would you get that kind of local flexibility in an example like this?
JED: Lets work this through as an example, because I think that by putting together the
two classes of examples that we had earlier, we can come up with something. First of
all, let us assume something that you didnt explicitly state: at some point the data from
the patient record is going to provide input to some less flexible system, like some kind
of health statistics record or something of that sort. So ultimately the fields maybe have
to be coded in order to adequately reflect the information. But in what order they get
filled in and how that coding eventually gets done is open. So to begin with, we could
provide the kind of margins that we were talking about earlier and allow, but not

To appear in Klaus Kaasgaard: Genres of Usability, forthcoming

require, that people can break out of this specific coding temporarily and move around
more flexibly. But then the issue becomes: How do we lock things back down? How do
we script it? And here it is important, I think, to realize that for the particular midwives
you talked to, they only had a choice between having it completely scripted or having it
open. But people might have, as Austin said, scripting needs that evolve over time. So
maybe for convenience, they might want to start on a different page or want to rearrange
what fields are on the page or what have you. Well, this is the sort of thing that you can
relatively easily do with HTML for example and maybe a little Java-script.
The realistic problem in this kind of situation is that most people wouldnt be able to do
that themselves. Its beyond what they should have to concern themselves with, its
beyond their skills. But in a social environment, typically you have people with
different skills and they pass little pieces around to each other, they help each other out
in various ways. So we would expect that if it was possible, if the basic material out of
which this application was made lends itself to this kind of social intervention, people
would adapt it. They would make friends that have similar needs and they would say:
Oh, thats just what I needed, could you give me a copy of that. Maybe the person
involved wouldnt even know how it works, but they can make a copy and give it to
you. So over time the base of software would evolve to fit the practices and not every
midwife would have the same things.
But then at the boundary you have to put filters, so that once a record reaches the
boundary, then if there is a mismatch or if the fields arent coded or whatever, it gets the
attention of the midwife, and says: Help. This wont go into the statistics database.
Please fill it in and then maybe the midwife goes through the process of translating the
annotations. So the system is requesting help from the user. Its a very flexible interplay
back and forth.
AUSTIN: And what you have just done in that last step is to introduce this idea of the
boundary; the boundary between before its gotten to the records processing and after.
But one can imagine other kinds of boundaries. So the idea that records will be sitting
within these spaces, which are connected to people in certain ways, and the idea that
that connection is the hook where the processing pieces could get applied, is something
which we typically dont do at the moment because we dont have those kinds of
boundaries. So this is the kind of invention purely within rigid computing that we need
to make. They are social inventions that need to get added to the conceptual structure of
rigid computing such that we can then allow this bricolage, this emergence and response
to the needs of the local circumstances.
JED: Maybe this is a good point to tie this back to the study of social discourses,
because what we are saying here is that the computer is not participating in the social
discourse. Its not an actor. But its enabling or supporting the discourse just the way
that pen and paper does, except just in a more rich and diverse way. Now we can start to
use the types of understandings that people use to analyze discourse: As an utterance or
some discursive material moves further and further from the community in which it was
generated, it tends to be reformulated in terms of broader and broader, more general
languages. So it gets translated from a very informal genre typically in a local
community to a more and more formal genre as it moves out into a larger community.

To appear in Klaus Kaasgaard: Genres of Usability, forthcoming

So there needs to be computational support for enabling and supporting and in some
case enforcing that translation. But its nothing exotic. Its something that fits into the
normal social process.
KLAUS: Im not sure that I hear you right. Are you saying that in this sense the
computer is a neutral tool? You said its not an actor in the social discourse.
JED: I dont mean that it is a neutral tool. I dont think that tools are ever neutral. They
always have a grain to them. But it is not a person. It doesnt, for example, deserve
respect or have rights in any personal sense. You dont expect it to make choices or
have intrinsic values that it would enforce in a situation. That is all I meant to say. It
isnt a social actor.
AUSTIN: Earlier I used the words the mechanics of something, in an attempt to
capture the idea that a lot of this stuff needs to be supported by putting mechanisms in
place by which stuff can get gathered together and bits of this processing can get
applied. Flags can be noticed and people can be called to take a look at this or that.
JED: People can be given an adequate context to decide what to do in a given case.
AUSTIN: Again, none of it is rocket science at that level.
KLAUS: Let me just stick with the example of the midwives. An obvious solution for
any usability specialist would be to provide the user with the ability to decide whether
she wanted this rigidity or whether she wanted to be able to move to another page
without having filled out the fields for instance. But this then gives us another problem,
a more traditional usability problem. The more flexibility you have, the more
complexity you typically have and the more difficult it is for people, who are not skilled
computer users, to use the system.
JED: Thats why we go back to the ecological approach. Most people would probably
copy something from a friend. If you brought all the complexity up to the surface and
there was an affordance for every decision, then that would be catastrophic. People only
want to see affordances for functions that they in fact use in practice. So it is always
necessary to be able to hide most of the possible choices below the surface. Right now
we hide them below the surface by having those choices made by the designers. The
designers do some analysis and bring the affordances up to the surface that they think
the community of users will need, or in some cases that the community of users should
be allowed to have. What we are saying is not that all the affordances should just be
floated up so that every user has to confront all possible choices, but that we should
accept the social re-construction of those choices over time. So instead of a designercontrolled interface, provide a tool kit, an environment in which users can build very
constrained sets of user interfaces and allow them to break those open and change the
choices. Most people wont break them open in fact, so you have to make it easy for
them to copy the ones they like. And then as part of that, you have to provide the
underlying infrastructure to map values across the different systems that are constructed.

To appear in Klaus Kaasgaard: Genres of Usability, forthcoming

AUSTIN: This actually is a common practice. Take UNIX systems. They always have
people breaking open and twiddling this file and that file. It gets completely out of their
hands; its hard to manage. If, on the other hand, you had an organized way of thinking
about all of that in a systematic way, then the process of breaking things open may
indeed be left to the individual user. You might start with certain things being openable
by end-users. And you might at some point say: Oh, the practice is changing such that
something which was two or three levels deep, now suddenly needs to be up where
people can get at it. So how does somebody reach down there and drag that up through
the intervening layers and how does that affect other people? That whole process is
beginning to move beyond the sort of thing that we know how to do in a straightforward
manner. There is some work to be done there.
JED: I just want to make the point that we are making processes here explicit which
arent ordinarily explicit. But in fact if you closely observe any moderately large group
of people, they are constantly engaged in negotiating these kinds of issues if the issues
havent somehow been locked down in software or otherwise taken out of their
control and even then they will often work around the attempts at control. People will
invent terminology, they will come to agreements on certain kinds of conventions, other
things will be left open and flexible. Theyll negotiate times close enough, theyll
establish routines and procedures in some cases and in other cases they wont. Then
perhaps the lack of a procedure becomes a problem or perhaps the existence of a
procedure becomes a problem. Sometimes things just change invisibly without
conscious effort, other times there is a crisis and people discuss it. All of these things
happen and what we are trying to do is to get the practice of software development to
integrate with that natural social process. We are not trying to create some new set of
skills that people dont have in the ordinary course of events.
AUSTIN: But having said that, we need to be careful because the usual practice of
software development is to think it all out in advance. This is true, even when you
iterate: things squeak, so you do a new specification, a new re-design and you roll it out.
Its piece-wise, in chunks, whereas what we would like to see is the honoring of what
happens in the real system, the social practices surrounding the rigid technical stuff,
and then take some of that and move it down and get the technology to support it. If you
can only do a re-design every six months, then what happens with those things that turn
up a month later, or didnt get into the last round? What do users do? Well, they figure
out something in the social world, which will handle it. Well put some stickies on, or
Ill write them down in my note-book and well remember. The system as a whole is of
course bubbling along; it doesnt stop changing. Current, the computing part of it may
not be bubbling along. All we are saying is: Lets admit to the fact that the world is
bubbling along, honor those practices and get computation to help a little.
JED: I would like to just mention that in talking to a lot of small Silicon Valley
companies, which are in the process of providing Web services, it is very clear that the
ones that are going to be successful have a practice of putting their stuff out early. They
are not trying to do a long design cycle. I have talked to people who are in successful
Web-service organizations like Amazon and their system evolves quite rapidly and
there is no real specification. The system is its own specification and there are many
parallel development efforts changing it at once. But they are all on very short cycles so

To appear in Klaus Kaasgaard: Genres of Usability, forthcoming

that they cant become detached from the main line of development for very long or
they become irrelevant. It is interesting also to watch open-source development projects
like Mozilla because they are generating new releases every night. There are many
different activities going on and they are all in open view, they are visible through email and bug reports and check-ins and there are exactly these kinds of social processes.
Sometimes things will break down or assumptions will fail, and then there will be
discussions and sometimes practices evolve. So there are many ways in which this is
actually visible in the software domain. I think actual software practices are now
outrunning the classical computer system design ideology quite nicely.
KLAUS: One of the problems here is that while all this sounds quite innocent, you are
actually asking people to give up power.
AUSTIN: Who?
KLAUS: Well, lets take the example with the midwives again. You are, in fact, asking
the health authorities to put more faith in the local interpretations, the local activities. So
I guess my questions could be formulated as this: Arent you giving up on the coherence
and promoting the local responsiveness on that behalf? And have you considered the
implications in terms of power structures?
AUSTIN: I guess that my reaction is that the idea that we have been able to maintain
the power of, say, the health authorities through the technology, that the technology has
been a device by which we limit the practice, a) may be an illusion because what is
really happening out there is not what actually happens in the technology, and b) is not a
necessary thing. That is to say that what you can enforce through the technology is not
the only thing that you can enforce. There are other social mechanisms. You can say to
people: Look, if you dont do it this way you lose your job or you lose your license.
Just because we cant do it through technology doesnt mean that we cannot do it. I
want to unload the burden that technology tends to have to bear of enforcing the law,
and say: Yep, its going to loosen that up and then the mechanisms for coherence, of
which enforcing the law may be one, are the subject matter of genuine debate rather
than a statement that this just has to be that way because that is the way the technology
is. I want to force that into the open.
KLAUS: My question would then be: Why should people, in this example the health
authorities, want to debate this?
JED: Right. I agree with Austin about this. I think there are some more things I want to
say along that line. But let me say, specifically, that the health authorities might not
want to debate it, but they arent autonomous either. They are embedded in the social
process as well. Standard sociological discourses on power to some extent seem to treat
power as somewhat autonomous. I wont go into the whole analysis but we see power as
being produced partly because people feel a need to structure organizations as a highly
coordinated system and someone ends up being able to take advantage of that
coordination. But even when the authorities are taking unfair advantage, people
tolerate the rigidity because in fact chaos really is worse. If you really ended up with

To appear in Klaus Kaasgaard: Genres of Usability, forthcoming

chaos it would be bad. People dont want that, so they are willing to pay quite a high
price to maintain the coordination.
However, if it is possible to have coordinated, well behaved social processes without the
rigidity, people will prefer that and theyll vote for it, in effect. They may vote
economically: these processes, for example, will tend to be more efficient; they may
vote simply by changing their membership in organizations; they may vote literally:
they may vote for a new government that chooses to allow more flexibility. But there
will be an overall social process that will change the rules.
Also, I think that there are other interesting things that Austin didnt mention about the
nature of this rigidity that we have today and the way technology tends to enforce it. To
a large extent the technology tends to put the burden, the cost of paying for this on the
people, say, at the shop floor, or in your case the midwives. Its not accounted for,
nobody counts up that cost. So it looks free, but it imposes penalties on the system as a
whole if, for example, the people on the shop floor have to work around the computer
system. In fact it takes more of their time, their productivity is lowered. But there is no
explicit visibility into that so management thinks that it is free. If they could see the
cost, if the costs became an explicit item and they were trading it off against higher
productivity, they would actually be thinking much more carefully about that.
So our goal in this is not simply to open everything up and make everything flexible.
Its to make the process of choosing a level of coherence and a tightness of coordination
visible and float that up close enough to the surface so that people can see the cost
trade-off people on the shop-floor and the management. On that background we can
then make reasonable trade-offs and negotiate as we do in our social situations and
come to some workable consensus. And we believe that the consensus has the potential
now to move considerably further toward flexibility because of a better technology.
AUSTIN: There are potentially some new points of equilibrium, some new patterns or
new practices of coherence that we havent explored yet. In fact, we might not even
know the language to begin to talk about this. We have this almost binary idea of either
having chaos or being locked down. Its either the military or its the market. And we
are saying that there is potentially lots of space in between, and there is a lot of work to
be done to understand that space. So if we were to spring, full blown on the world, a
technology that could do all the things that we advocate either the simpler ones or the
much richer ones we would be confronted with the fact that we wouldnt know how to
talk about it. We would have to develop those techniques. And one of the stable points
in this very rich technology, that we imagine, is the military style of thinking where
everything is locked down. And if people want to go that route then they can still go that
route. Its not as if you have the rigid systems on one hand and the pliant systems on the
other. Rather, pliant systems include the rigid systems as a special case.
JED: That is an option. I think locked down has been stable historically, but I think it
is becoming unstable, and I actually think you can see this in a very macroscopic sense
with the positive terror that copier technology and now the Internet inspires in
totalitarian regimes. They see right away that this is the death of their ability to retain

To appear in Klaus Kaasgaard: Genres of Usability, forthcoming

control. And I think they feel terror for this reason. So they try to suppress these
technologies, but I think most of us believe they wont succeed.
KLAUS: Lets move more in the direction of your second solution which requires
much more research than what we have been discussing up to now. Could you expand a
little bit on what technologies are required to enable such systems? Also, I would like to
have you expand more on what social forces will oppose and encourage them? These
are large questions, so let's start with what technologies are required to enable pliant
systems.
JED: Let me just provide a little background. We see that we can go a long way and we
dont know how far with the pliant design of systems that use todays rigid technology.
But ultimately that runs into inherent limitations in the technology itself. For example,
computers tend to be based on very discrete values. I dont mean discrete just on the bit
level, but the whole structure of computer programs is sequential and you end up
making lots of discrete decisions. And the problem is that this kind of discrete choice is
very unforgiving of any mistakes.
We believe that ultimately, in order to get full value from the pliant perspective, we will
have to go to more radical approaches that will change the nature of computation itself.
And right now there are technologies that do this such as neural networks and various
kinds of Bayesian computations that are used in robotics. So we are not talking about
something that nobody has ever seen before. The problem today is that those
technologies are only useful in very restricted areas, like for example hand writing
recognition or speech recognition or maybe some kind of process to decide whether a
screw is defective based on visual inspection. They arent really computational
processes in the broad sense. You cant build big, complex systems out of them. They
dont scale. What weve tried to do is to imagine what it would take to make those much
more scaleable.
Essentially, the key thing that I believe is missing from these softer technologies
today is compositionality. There are a lot of different names for it, but basically the
wonderful thing about human language and even more perhaps algebraic languages
is that you can take pieces and bind them together. This is of course also a key attribute
of computer systems, both the hardware and the software. You can plug pieces together
and they somehow combine their functions and if you do it in an appropriate way then
the whole is greater than the sum of the parts. Right now, if you have a whole bunch of
different neural networks, there is really no useful way to plug them together in the
sense where you get a bigger, more complex, pliant system out of them.
There are a few examples, though, of compositionality for pliant systems. We think that
one of the really interesting examples of that kind of compositionality, which appears to
be pliant to us, is Christopher Alexanders pattern languages for architectural design.iii I
want to make a distinction here. We dont mean the kind of patterns that have been used
in software design. Alexanders approach is really much more flexible and continuous
than the typical software patterns. So what we are interested in doing is finding ways to
put Alexanders pattern languages on a computational footing. Not in the sense that
computers would somehow gain the abilities that architects have using this language,

To appear in Klaus Kaasgaard: Genres of Usability, forthcoming

but just in the sense that computers would vivify those patterns, make them alive and
responsive. So when you build something with those patterns, it would be active and
responsive the way things can be when you build them out of code.
AUSTIN: One of the components of active patterns, which really makes a sharp
distinction with object oriented programming patterns, is the act of creating an instance
of some pattern, or an instance of a class. In object oriented programming such an act in
no way threatens to change the class or the conception that you are instantiating. The
nature of the pattern language that we are imagining, is that the act of seeing some
situation as a case of a particular pattern, an act which takes work to do, may in fact
enrich your notion of what that pattern is and may cause you to adjust the pattern itself.
So the very act of trying to think about something as a case of something youve seen
before, threatens your understanding of what you had before. Now, that is terrifying to
those who want to know what they have got. So again we come back to the question of
how you are going to keep it coherent. But the technology, at the fundamental level of
moment by moment seeing something as a case of a pattern, is already beginning to
move, which then affects other things. And its not that you are just seeing something as
one pattern. You are seeing it in a structure of patterns so the process of bashing things
together is happening not in the abstract, nothing happens much in the abstract. It
happens in the very specifics of doing a particular thing. Let me give an example: In
1978, Eleanor Wynn observed a Xerox clerk taking phone orders for copiers supplies
(paper and toner).iv As part of taking orders, the clerk got a shipping address from the
customer. One customer had trouble providing the shipping address, because the copier
was on an ocean-going barge: if Xerox could say when the supplies would be shipped,
the customer could say where to ship them. The form wanted an address; the situation
could not produce one. At this moment, Xerox (in the person of the order clerk) faced a
conceptual shift: Oops I see, an address can be time varying. I hadnt thought of that.
The act of confronting this clerk with an address which was time varying caused the
pattern which said you gotta get the shipping address to confront the fact that the
address in this particular case is going to be time varying. How are we going to do that?
So the pattern itself had to be extended to deal with the situation.
JED: I want to point out an implicit theme here. We are not talking about introducing
some radical new technology, which is going to do some wonderful thing. We are
talking about making it easier for people to do what they are doing anyway. This kind of
pattern, the seeing-as, the incremental reshaping of practices through their application in
particular cases is something that happens in the social context. Right now computers
are very rigid. We are talking about making them easier to modify to begin with and
then we are talking about incrementally over time developing technologies that make it
easier and easier for the social process to reshape the computational process. We are
also saying that in parallel that it is necessary to develop a better understanding of how
the social process controls itself, how it controls the coherence of its structures and
develop a taxonomy, a language for talking about it. And that will itself indirectly
develop into computational techniques. So the reason we talk about the long term
picture the more continuous computation and Alexandrian patterns is not so much to
say: Well, lets just start working on that and create some autonomous technology,
which we can then throw into the social process, because it is not going to happen that
way. It is to give us a vision for how this will develop over a longer period. And sure,

To appear in Klaus Kaasgaard: Genres of Usability, forthcoming

we can do the technology development, but it will have to be constantly re-integrated in

the social process. The two will have to develop together.
KLAUS: Let me just step back a little. Before we end, I would like to take a really
pragmatic look at this and see it from, for instance, my point of view or my companys
point of view. What would be your advice to me and to companies like KMD? In many
ways, I think a lot of people will agree with you: The midwives that we talked about
before would probably agree with you (if they had any idea of what we are talking
about) and I as a usability specialist agree with some of your fundamental principles. If
this sells better, people at management-level in my company will also agree with you.
But what should we do?
JED: Well, one thing you could do and you have a more precise sense of how to
execute this is to develop a tool-kit for creating these applications of the sort we were
talking about in the midwife example: a software infrastructure on which you could put
these fragments together and allow the user community to copy them and recombine
them on their own and you can lock down some of them. Underneath that there has to
be a database and probably a conventional database would be adequate as long as it can
store comments as well as other pieces of data.
AUSTIN: Even before you did that, there is the need to expose people who are building
systems to this whole concept and the possibility of pliant systems and then letting them
say: Okay, Im going to respond to it by adding a place for the margin, for example.
JED: I agree. Change the discourse first.
AUSTIN: Exactly. Get that going and then look at the regular patterns that are
beginning to emerge out of that. Youve got some really smart people. They are going
to begin to see regularities: Oh, I keep adding the same thing. Good, thats what the
toolkit should do. Then let that follow. We can guess what some of these things might
be, but from a pure practical point of view, if you could infuse people with the spirit of
the thing, then you would be getting them to begin to design these things.
Correspondingly, you would be having all those discussions with your customers, with
the midwives and others, about what is actually needed. Is there a need for a new kind
of specialist in the field etc.? That needs to be constructed because you have been
arguing that you need not only the technical stuff but also the social stuff that goes with
it.
KLAUS: One last thing: Could you think of or have you actually met social forces
that oppose these ideas?
JED: Well, I think there are a lot of social forces that oppose these ideas at many
different levels. In the United States there is a very strong social tradition, which
interestingly is most clearly represented in religious thought, things like biblical
inerrancy and opposition to teaching critical thinking and a very rigid idea of truth and
falsehood. I think that forces like that very broadly based social forces will oppose
this. I think that management in some companies will oppose this because of fear of loss
of control. But I am not particularly concerned about that because I think the economic

To appear in Klaus Kaasgaard: Genres of Usability, forthcoming

and social forces will simply go around those companies and they will become obsolete
and go out of business or change their policies. In a relatively free market that kind of
process is pretty effective. I think that people who are socialized to existing practices of
system design will probably be very uncomfortable, because this is not an easy
transition for people to make.
AUSTIN: We have seen that regularly just in giving talks. People have trouble getting
the idea of what it would be like.
JED: I think that some of these people will have a much easier time once there are more
working examples. But other people will just fundamentally dislike it because they like
very precise, crisp forms that can be combined in well defined ways, and they basically
want to work in an environment where things are neat, clean and under control.
KLAUS: Imagine I was trained in computer science and spent 15 years learning these
programming languages that are quite hard to learn because they are so detached from
real-life problems. That gives me a certain position and in certain places of the world,
like here in Silicon Valley, it enables me to write my own paycheck because
programmers are hard to get these days. So I could imagine you would have some
opposition there too.
JED: Thats right. Information systems professionals profit from the obscurity of their
technology. If you give more power to the users, then it undercuts this. So there are lots
of constituencies that will oppose this.
AUSTIN: You will also find the deep social forces that Jed was talking about not only
in the United States but in the whole western tradition of this idea of the one. A signal is
the use of the articles the and a. The moment you say the you are in that position
of getting one answer. This is opposed to our view, which is that we need to have a
coherent set of ways of looking at the world which can work together rather than
getting the solution. It is so deeply built into the way we think. Most of what we have
been struggling with in doing this work for four years now, has been confronting those
assumptions in ourselves even when we were consciously aware that this is what we are
trying to do.
KLAUS: Have you experienced any differences between speaking about this here in the
States and speaking about it in Scandinavia? Scandinavian countries have a history of
taking different perspectives quite seriously in debating things before reaching
consensus, probably more than in the States?
AUSTIN: I have given the talk in Denmark twice, once at Aarhus University and once
at Danfossv, and I found a resonance. On the other hand, we have found a resonance
most of the places that we have given the talk here. But we have also been very
selective in finding those places where we thought there would be resonance. I think
that there is somewhat less resonance in North America but we are talking to those who
are interested in HCI or have this view that the unit of analysis is the socio-technical
practice and that resonates.

To appear in Klaus Kaasgaard: Genres of Usability, forthcoming

JED: I have only given this talk in the United States so I cant make the contrast. But I
think that the biggest issue weve seen in talking about this is difficulty in getting
people to overcome assumptions that simply make this idea incomprehensible. I dont
think the problem is that we have got people who understand it but push back and dont
like it. I think the problem is a failure of imagination. Once we can get people over that
failure of imagination we often find that they get actively interested. So I think it is
partly our own failure but I think it is largely an indication of how hard it is to get past
the underlying assumptions that are so deeply built into our culture.
KLAUS: Well, this talk has certainly helped me understand some of these ideas. Thank
you very much both of you.

To appear in Klaus Kaasgaard: Genres of Usability, forthcoming

Joanne Yates: Control Through Communication: The Rise of System in American Management, John
Hopkins University Press, Reprint Edition, 1993.
ii

Xeroxs European Research Center in Cambridge, England.

iii

Christopher Alexanders core books on architectural design and pattern languages are:
Christopher Alexander: Notes on the Synthesis of Form, Harvard University Press, 1964.
Christopher Alexander, Sara Ishikawa & Murray Silverstein: A Pattern Language: Towns, Buildings,
Construction, Oxford University Press, 1977.
Christopher Alexander: The Timeless Way of Building, Oxford University Press, 1979.
iv

Elanor Wynn: Office Conversation as an Information Medium. Unpublished Ph.D. thesis, University
of California, Berkeley, 1979. For a paper that uses Wynns observations as a centeral driver to discuss
the difficulty of supporting office procedures, see: See Austin Henderson & R. E. Fikes: On Supporting
the Use of Procedures in Office Work, in Proceedings of the First Annual National Conference on
Artificial Intelligence, American Association of Artificial Intelligence, Menlo Park, California, 1980.
v

Danfoss is Denmark's largest industrial group with about 20,000 employees. Danfoss develops and
produces mechanical and electronic components for several industrial branches worldwide.